Switching data-processing methods in a control loop: Trade-off between delay and probability of data acquisition

Many control applications, such as vision-based control, require data-processing methods to distill sensor information. This data-processing introduces several undesired effects in the control loop, such as delays, the probability of not acquiring information, and measurement inaccuracies. Often, these effects obey a trade-off. For example, the probability of acquiring control-relevant information, related to the probability of data-loss, is typically higher if a larger delay is allowed. While a single processing method with a reasonable trade-off is typically selected, we propose instead a solution to switch between data-processing methods with different delays and corresponding data-loss probabilities. We prove that the proposed method achieves a better LQG-type performance when compared to the individual methods. A simulation considering a second-order system illustrates the advantages of the proposed method.